Slides on my website

https://phdenzel.github.io/

Outlook

• 2D galaxy modelling
• 3D modelling using point clouds
  • First experiments
• 6D phase-space modelling?

Recap:
Generative models
for map-to-map translation

Dataset from IllustrisTNG

• projected IllustrisTNG galaxies
• 6 domains:
  • dark-matter, stars, gas,
    HI, temperature, magnetic field
• ∼ 2'000 galaxies, (across 6 snapshots)
• ∼ 360'000 images
• each galaxy \(\ge\) 10'000 particles
• augmented: up to 5x randomly rotated
• scale: 2 dark-matter half-mass radii

Dataset from IllustrisTNG

• projected IllustrisTNG galaxies
• 6 domains:
  • dark-matter, stars, gas,
    HI, temperature, magnetic field
• ∼ 2'000 galaxies, (across 6 snapshots)
• ∼ 360'000 images
• each galaxy \(\ge\) 10'000 particles
• augmented: up to 5x randomly rotated
• scale: 2 dark-matter half-mass radii

Generative model architectures

Benchmark of generative models we're investigating and comparing:

• cGANs: lackluster results (see previous talk)
• Score-based diffusion models: promising results but really slow
• InDI models: more efficient at inference?
• Diffusion Mamba: the latest and greatest?

cGANs

Score-based diffusion (SDM)

Noise schedule

Inversion by Direct Iteration (InDI)

Diffusion Mamba (DiM)

From 2D to 3D models

• observations inherently have 2D spatial resolution
• astrophysical structures are inherently 3D
• modelling difficulties:
  • inherent 3D features, different 2D perspectives
  • degeneracies

Inherent 3D shapes

Degeneracies

Point-cloud models

Point-cloud experiments

"Phase-space-point" models

References

• SDM: Song et al. (2021)
• InDI: Delbracio & Milanfar (2023)
• DiM: Teng et al. (2024)
• PointNet for summary statistics: Anagnostidis et al. (2022)
• Point cloud generation for galaxy surveys: Cuesta-Lazaro & Mishra-Sharma (2023)
• Cosmology from point clouds: Chatterjee & Villaescusa-Navarro (2024)